This paper proposes a power grid fault diagnosis method based on cluster analysis. The method uses multi-source fault information to determine faulty components, and it can get accurately result in the absence of information. The multi-source fault information used in this paper includes wide-area system measurement information, fault recorder information and protection action information. The method firstly uses multi-level information analysis method to analyze the power grid fault information, extracts the fault features, and determines the suspicious fault component set through the clustering result and the switch action information. Then using the improved Bayesian probability model to fuse fault information and identify faulty components. Finally, this paper verifies the effectiveness and feasibility of the proposed method through concrete examples.